--- trunk/mkinitrd-magellan/busybox/archival/libunarchive/decompress_unlzma.c	2009/04/24 18:32:46	815
+++ trunk/mkinitrd-magellan/busybox/archival/libunarchive/decompress_unlzma.c	2009/04/24 18:33:46	816
@@ -12,8 +12,8 @@
 #include "libbb.h"
 #include "unarchive.h"
 
-#ifdef CONFIG_FEATURE_LZMA_FAST
-#  define speed_inline ATTRIBUTE_ALWAYS_INLINE
+#if ENABLE_FEATURE_LZMA_FAST
+#  define speed_inline ALWAYS_INLINE
 #else
 #  define speed_inline
 #endif
@@ -45,7 +45,7 @@
 
 
 /* Called twice: once at startup and once in rc_normalize() */
-static void rc_read(rc_t * rc)
+static void rc_read(rc_t *rc)
 {
 	int buffer_size = safe_read(rc->fd, RC_BUFFER, RC_BUFFER_SIZE);
 	if (buffer_size <= 0)
@@ -58,9 +58,9 @@
 static rc_t* rc_init(int fd) /*, int buffer_size) */
 {
 	int i;
-	rc_t* rc;
+	rc_t *rc;
 
-	rc = xmalloc(sizeof(rc_t) + RC_BUFFER_SIZE);
+	rc = xmalloc(sizeof(*rc) + RC_BUFFER_SIZE);
 
 	rc->fd = fd;
 	/* rc->buffer_size = buffer_size; */
@@ -78,50 +78,51 @@
 }
 
 /* Called once  */
-static ATTRIBUTE_ALWAYS_INLINE void rc_free(rc_t * rc)
+static ALWAYS_INLINE void rc_free(rc_t *rc)
 {
-	if (ENABLE_FEATURE_CLEAN_UP)
-		free(rc);
+	free(rc);
 }
 
 /* Called twice, but one callsite is in speed_inline'd rc_is_bit_0_helper() */
-static void rc_do_normalize(rc_t * rc)
+static void rc_do_normalize(rc_t *rc)
 {
 	if (rc->ptr >= rc->buffer_end)
 		rc_read(rc);
 	rc->range <<= 8;
 	rc->code = (rc->code << 8) | *rc->ptr++;
 }
-static ATTRIBUTE_ALWAYS_INLINE void rc_normalize(rc_t * rc)
+static ALWAYS_INLINE void rc_normalize(rc_t *rc)
 {
 	if (rc->range < (1 << RC_TOP_BITS)) {
 		rc_do_normalize(rc);
 	}
 }
 
-/* Called 9 times */
+/* rc_is_bit_0 is called 9 times */
 /* Why rc_is_bit_0_helper exists?
- * Because we want to always expose (rc->code < rc->bound) to optimizer
+ * Because we want to always expose (rc->code < rc->bound) to optimizer.
+ * Thus rc_is_bit_0 is always inlined, and rc_is_bit_0_helper is inlined
+ * only if we compile for speed.
  */
-static speed_inline uint32_t rc_is_bit_0_helper(rc_t * rc, uint16_t * p)
+static speed_inline uint32_t rc_is_bit_0_helper(rc_t *rc, uint16_t *p)
 {
 	rc_normalize(rc);
 	rc->bound = *p * (rc->range >> RC_MODEL_TOTAL_BITS);
 	return rc->bound;
 }
-static ATTRIBUTE_ALWAYS_INLINE int rc_is_bit_0(rc_t * rc, uint16_t * p)
+static ALWAYS_INLINE int rc_is_bit_0(rc_t *rc, uint16_t *p)
 {
 	uint32_t t = rc_is_bit_0_helper(rc, p);
 	return rc->code < t;
 }
 
 /* Called ~10 times, but very small, thus inlined */
-static speed_inline void rc_update_bit_0(rc_t * rc, uint16_t * p)
+static speed_inline void rc_update_bit_0(rc_t *rc, uint16_t *p)
 {
 	rc->range = rc->bound;
 	*p += ((1 << RC_MODEL_TOTAL_BITS) - *p) >> RC_MOVE_BITS;
 }
-static speed_inline void rc_update_bit_1(rc_t * rc, uint16_t * p)
+static speed_inline void rc_update_bit_1(rc_t *rc, uint16_t *p)
 {
 	rc->range -= rc->bound;
 	rc->code -= rc->bound;
@@ -129,7 +130,7 @@
 }
 
 /* Called 4 times in unlzma loop */
-static int rc_get_bit(rc_t * rc, uint16_t * p, int *symbol)
+static int rc_get_bit(rc_t *rc, uint16_t *p, int *symbol)
 {
 	if (rc_is_bit_0(rc, p)) {
 		rc_update_bit_0(rc, p);
@@ -143,7 +144,7 @@
 }
 
 /* Called once */
-static ATTRIBUTE_ALWAYS_INLINE int rc_direct_bit(rc_t * rc)
+static ALWAYS_INLINE int rc_direct_bit(rc_t *rc)
 {
 	rc_normalize(rc);
 	rc->range >>= 1;
@@ -156,7 +157,7 @@
 
 /* Called twice */
 static speed_inline void
-rc_bit_tree_decode(rc_t * rc, uint16_t * p, int num_levels, int *symbol)
+rc_bit_tree_decode(rc_t *rc, uint16_t *p, int num_levels, int *symbol)
 {
 	int i = num_levels;
 
@@ -227,8 +228,8 @@
 };
 
 
-USE_DESKTOP(long long) int
-unlzma(int src_fd, int dst_fd)
+USE_DESKTOP(long long) int FAST_FUNC
+unpack_lzma_stream(int src_fd, int dst_fd)
 {
 	USE_DESKTOP(long long total_written = 0;)
 	lzma_header_t header;
@@ -280,14 +281,16 @@
 	while (global_pos + buffer_pos < header.dst_size) {
 		int pos_state = (buffer_pos + global_pos) & pos_state_mask;
 
-		prob =
-			p + LZMA_IS_MATCH + (state << LZMA_NUM_POS_BITS_MAX) + pos_state;
+		prob = p + LZMA_IS_MATCH + (state << LZMA_NUM_POS_BITS_MAX) + pos_state;
 		if (rc_is_bit_0(rc, prob)) {
 			mi = 1;
 			rc_update_bit_0(rc, prob);
-			prob = (p + LZMA_LITERAL + (LZMA_LIT_SIZE
-					* ((((buffer_pos + global_pos) & literal_pos_mask) << lc)
-					+ (previous_byte >> (8 - lc)))));
+			prob = (p + LZMA_LITERAL
+			        + (LZMA_LIT_SIZE * ((((buffer_pos + global_pos) & literal_pos_mask) << lc)
+			                            + (previous_byte >> (8 - lc))
+			                           )
+			          )
+			);
 
 			if (state >= LZMA_NUM_LIT_STATES) {
 				int match_byte;
@@ -302,35 +305,37 @@
 					match_byte <<= 1;
 					bit = match_byte & 0x100;
 					prob_lit = prob + 0x100 + bit + mi;
-					if (rc_get_bit(rc, prob_lit, &mi)) {
-						if (!bit)
-							break;
-					} else {
-						if (bit)
-							break;
-					}
+					bit ^= (rc_get_bit(rc, prob_lit, &mi) << 8); /* 0x100 or 0 */
+					if (bit)
+						break;
 				} while (mi < 0x100);
 			}
 			while (mi < 0x100) {
 				prob_lit = prob + mi;
 				rc_get_bit(rc, prob_lit, &mi);
 			}
-			previous_byte = (uint8_t) mi;
 
+			state -= 3;
+			if (state < 4-3)
+				state = 0;
+			if (state >= 10-3)
+				state -= 6-3;
+
+			previous_byte = (uint8_t) mi;
+#if ENABLE_FEATURE_LZMA_FAST
+ one_byte1:
 			buffer[buffer_pos++] = previous_byte;
 			if (buffer_pos == header.dict_size) {
 				buffer_pos = 0;
 				global_pos += header.dict_size;
-				if (full_write(dst_fd, buffer, header.dict_size) != header.dict_size)
+				if (full_write(dst_fd, buffer, header.dict_size) != (ssize_t)header.dict_size)
 					goto bad;
 				USE_DESKTOP(total_written += header.dict_size;)
 			}
-			if (state < 4)
-				state = 0;
-			else if (state < 10)
-				state -= 3;
-			else
-				state -= 6;
+#else
+			len = 1;
+			goto one_byte2;
+#endif
 		} else {
 			int offset;
 			uint16_t *prob_len;
@@ -350,24 +355,23 @@
 				if (rc_is_bit_0(rc, prob)) {
 					rc_update_bit_0(rc, prob);
 					prob = (p + LZMA_IS_REP_0_LONG
-							+ (state << LZMA_NUM_POS_BITS_MAX) + pos_state);
+					        + (state << LZMA_NUM_POS_BITS_MAX)
+					        + pos_state
+					);
 					if (rc_is_bit_0(rc, prob)) {
 						rc_update_bit_0(rc, prob);
 
 						state = state < LZMA_NUM_LIT_STATES ? 9 : 11;
+#if ENABLE_FEATURE_LZMA_FAST
 						pos = buffer_pos - rep0;
 						while (pos >= header.dict_size)
 							pos += header.dict_size;
 						previous_byte = buffer[pos];
-						buffer[buffer_pos++] = previous_byte;
-						if (buffer_pos == header.dict_size) {
-							buffer_pos = 0;
-							global_pos += header.dict_size;
-							if (full_write(dst_fd, buffer, header.dict_size) != header.dict_size)
-								goto bad;
-							USE_DESKTOP(total_written += header.dict_size;)
-						}
-						continue;
+						goto one_byte1;
+#else
+						len = 1;
+						goto string;
+#endif
 					} else {
 						rc_update_bit_1(rc, prob);
 					}
@@ -403,7 +407,7 @@
 			if (rc_is_bit_0(rc, prob_len)) {
 				rc_update_bit_0(rc, prob_len);
 				prob_len = (prob + LZMA_LEN_LOW
-							+ (pos_state << LZMA_LEN_NUM_LOW_BITS));
+				            + (pos_state << LZMA_LEN_NUM_LOW_BITS));
 				offset = 0;
 				num_bits = LZMA_LEN_NUM_LOW_BITS;
 			} else {
@@ -412,14 +416,14 @@
 				if (rc_is_bit_0(rc, prob_len)) {
 					rc_update_bit_0(rc, prob_len);
 					prob_len = (prob + LZMA_LEN_MID
-								+ (pos_state << LZMA_LEN_NUM_MID_BITS));
+					            + (pos_state << LZMA_LEN_NUM_MID_BITS));
 					offset = 1 << LZMA_LEN_NUM_LOW_BITS;
 					num_bits = LZMA_LEN_NUM_MID_BITS;
 				} else {
 					rc_update_bit_1(rc, prob_len);
 					prob_len = prob + LZMA_LEN_HIGH;
 					offset = ((1 << LZMA_LEN_NUM_LOW_BITS)
-							  + (1 << LZMA_LEN_NUM_MID_BITS));
+					          + (1 << LZMA_LEN_NUM_MID_BITS));
 					num_bits = LZMA_LEN_NUM_HIGH_BITS;
 				}
 			}
@@ -430,12 +434,10 @@
 				int pos_slot;
 
 				state += LZMA_NUM_LIT_STATES;
-				prob =
-					p + LZMA_POS_SLOT +
-					((len <
-					  LZMA_NUM_LEN_TO_POS_STATES ? len :
-					  LZMA_NUM_LEN_TO_POS_STATES - 1)
-					 << LZMA_NUM_POS_SLOT_BITS);
+				prob = p + LZMA_POS_SLOT +
+				       ((len < LZMA_NUM_LEN_TO_POS_STATES ? len :
+				         LZMA_NUM_LEN_TO_POS_STATES - 1)
+				         << LZMA_NUM_POS_SLOT_BITS);
 				rc_bit_tree_decode(rc, prob, LZMA_NUM_POS_SLOT_BITS,
 								   &pos_slot);
 				if (pos_slot >= LZMA_START_POS_MODEL_INDEX) {
@@ -466,17 +468,18 @@
 			}
 
 			len += LZMA_MATCH_MIN_LEN;
-
+ SKIP_FEATURE_LZMA_FAST(string:)
 			do {
 				pos = buffer_pos - rep0;
 				while (pos >= header.dict_size)
 					pos += header.dict_size;
 				previous_byte = buffer[pos];
+ SKIP_FEATURE_LZMA_FAST(one_byte2:)
 				buffer[buffer_pos++] = previous_byte;
 				if (buffer_pos == header.dict_size) {
 					buffer_pos = 0;
 					global_pos += header.dict_size;
-					if (full_write(dst_fd, buffer, header.dict_size) != header.dict_size)
+					if (full_write(dst_fd, buffer, header.dict_size) != (ssize_t)header.dict_size)
 						goto bad;
 					USE_DESKTOP(total_written += header.dict_size;)
 				}
@@ -485,13 +488,16 @@
 		}
 	}
 
-
-	if (full_write(dst_fd, buffer, buffer_pos) != buffer_pos) {
+	{
+		SKIP_DESKTOP(int total_written = 0; /* success */)
+		USE_DESKTOP(total_written += buffer_pos;)
+		if (full_write(dst_fd, buffer, buffer_pos) != (ssize_t)buffer_pos) {
  bad:
+			total_written = -1; /* failure */
+		}
 		rc_free(rc);
-		return -1;
+		free(p);
+		free(buffer);
+		return total_written;
 	}
-	rc_free(rc);
-	USE_DESKTOP(total_written += buffer_pos;)
-	return USE_DESKTOP(total_written) + 0;
 }